The automotive and agriculture industries are experiencing rapid changes in global standards and customer expectations. This presents challenges for equipment manufacturers and tier-one suppliers who must develop and continuously improve complex systems by considering a wide range of user operations, while also prioritizing low cost, reduced emissions, and high reliability of the final products. The after-sales phase of the product lifecycle involves high-cost risks, as failures can lead to warranty claims and impact overall customer satisfaction. Modern off-highway machines are equipped with connectivity devices and sensors that can send selected information to the cloud environment of the vehicle manufacturer. This data can be analyzed and interpreted to provide solutions to the challenges faced in the after-market business field. This paper proposes a methodology to leverage a massive amount of data collected by a fleet of agricultural machines spread across Europe to support dealers in scheduling recurring services. The machine learning-based methodology aims at predicting the extent to which the vehicle will be operated in the upcoming days and weeks, allowing for the next service date prediction, and consequently supporting dealers in scheduling recurring services. The approach involves collecting telemetry data coming from the vehicles, combining this dataset with historical weather data recorded by surrounding weather stations, performing feature engineering, and extracting operating profiles to finally predict the future vehicle behavior and next service date. The performance of the final model was evaluated using different metrics. A benchmarking model was also considered to better assess the quality of the results obtained with a machine-learning regression model. The results showed that the machine learning model outperformed the baseline model, and the prediction could provide a valuable indication to support the dealer in planning the service and taking all needed actions in advance. An identified limitation of this analysis was the lack of information about the type of farming performed and the extension of the area covered by each agricultural vehicle. This limitation could partially be overcome by the data-driven usage profile analysis. However, more precise information could potentially improve the accuracy of the forecast. To the best of the authors' knowledge, no similar analysis has been conducted in the literature using such a large amount of field data. The quality of the obtained results led to the deployment and integration of the model into the OEMs Customer-Relationship-Management (CRM) system, where it is used to predict the next service date and presented to dealers to efficiently plan workload and resources.
Dipl.-Ing. Chiara Gei, Data Scientist, AVL List GmbH